An observer study of reconstruction strategies for the detection of solitary pulmonary nodules using hybrid NeoTect SPECT images

Abstract

The optimization of reconstruction parameters, and determination of whether iterative reconstruction should replace FBP clinically should be based on tasks which closely approximate the clinical application of the images. The use of hybrid images or studies in which simulated lesions are added to known normal clinical acquisitions represents a practical alternative to the use of purely clinical acquisitions for observer studies. The goal of this investigation was to use hybrid images of the lung tumor imaging agent Tc-99m NeoTect in localization receiver operating characteristic (LROC) studies to determine reconstruction parameters and whether iterative reconstruction with attenuation, scatter, and distance-dependent resolution compensation should replace FBP clinically. Nine patient's clinically normal NeoTect SPECT images were used in this study. The lung lesions were created by adding Monte Carlo simulated tumor projections to the patient's acquisition projections. All tumors were spheres 1 cm diameter in size which is the smallest tumor could be identified by CT. Attenuation, scatter and distance dependent resolution compensations were all included in the RBI-EM reconstruction. The iteration number tested were 1, 3, 5, 7 and 10, and the Gaussian postfilter's FWHM tested were 0, 1, 2, 3, and 4 image pixels. The Butterworth filter's cutoff frequencies 0.10, 0.15, 0.20, 0.25 and 0.30 (1/pixel) were tested for the FBP reconstruction. A total of 324 images were produced for each of the reconstruction strategies, with half of them are normal images. Channelized nonprewhitening (CNPW) numerical and human observers were employed to investigate tumor detection accuracy. The RBI-EM with attenuation, scatter and distance dependent resolution compensations performed better than the FBP in lung tumor detection and localization. The best performance (highest score of the area under the LROC curves) was for 5 iterations and 1 pixel's FWHM of Gaussian postfiltering in the RBI-EM reconstruction.